Cluster Accurate Scheduler Estimator For Rescheduling

Users could divide their replicas of a workload into different clusters in terms of available resources of member clusters. When some clusters are lack of resources, scheduler would not assign excessive replicas into these clusters by calling karmada-scheduler-estimator.

Prerequisites

Karmada has been installed

We can install Karmada by referring to quick-start, or directly run hack/local-up-karmada.sh script which is also used to run our E2E cases.

Member cluster component is ready

Ensure that all member clusters have been joined and their corresponding karmada-scheduler-estimator is installed into karmada-host.

You could check by using the following command:

# check whether the member cluster has been joined
$ kubectl get cluster
NAME       VERSION   MODE   READY   AGE
member1    v1.19.1   Push   True    11m
member2    v1.19.1   Push   True    11m
member3    v1.19.1   Pull   True    5m12s

# check whether the karmada-scheduler-estimator of a member cluster has been working well
$ kubectl --context karmada-host get pod -n karmada-system | grep estimator
karmada-scheduler-estimator-member1-696b54fd56-xt789   1/1     Running   0          77s
karmada-scheduler-estimator-member2-774fb84c5d-md4wt   1/1     Running   0          75s
karmada-scheduler-estimator-member3-5c7d87f4b4-76gv9   1/1     Running   0          72s

• If you deploy Karmada in Physical Machine, see Installation by Binary.
• If you deploy Karmada in Host Kubernetes Cluster.
  • If the Member Cluster has not been joined, you could use hack/deploy-agent-and-estimator.sh to deploy both karmada-agent and karmada-scheduler-estimator.
  • If the Member Cluster has been joined already, you could use hack/deploy-scheduler-estimator.sh to only deploy karmada-scheduler-estimator.

Scheduler option '--enable-scheduler-estimator'

After all member clusters have been joined and estimators are all ready, please specify the option --enable-scheduler-estimator=true to enable scheduler estimator.

# edit the deployment of karmada-scheduler
kubectl --context karmada-host edit -n karmada-system deployments.apps karmada-scheduler

And then add the option --enable-scheduler-estimator=true into the command of container karmada-scheduler.

Example

Now we could divide the replicas into different member clusters. Note that propagationPolicy.spec.replicaScheduling.replicaSchedulingType must be Divided and propagationPolicy.spec.replicaScheduling.replicaDivisionPreference must be Aggregated. The scheduler will try to divide the replicas aggregately in terms of all available resources of member clusters.

apiVersion: policy.karmada.io/v1alpha1
kind: PropagationPolicy
metadata:
  name: aggregated-policy
spec:
  resourceSelectors:
    - apiVersion: apps/v1
      kind: Deployment
      name: nginx
  placement:
    clusterAffinity:
      clusterNames:
        - member1
        - member2
        - member3
    replicaScheduling:
      replicaSchedulingType: Divided
      replicaDivisionPreference: Aggregated

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx
  labels:
    app: nginx
spec:
  replicas: 5
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - image: nginx
        name: nginx
        ports:
        - containerPort: 80
          name: web-1
        resources:
          requests:
            cpu: "1"
            memory: 2Gi

You will find all replicas have been assigned to as few clusters as possible.

$ kubectl get deployments.apps          
NAME    READY   UP-TO-DATE   AVAILABLE   AGE
nginx   5/5     5            5           2m16s
$ kubectl get rb nginx-deployment -o=custom-columns=NAME:.metadata.name,CLUSTER:.spec.clusters  
NAME               CLUSTER
nginx-deployment   [map[name:member1 replicas:5] map[name:member2] map[name:member3]]

After that, we change the resource request of the deployment to a large number and have a try again.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx
  labels:
    app: nginx
spec:
  replicas: 5
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - image: nginx
        name: nginx
        ports:
        - containerPort: 80
          name: web-1
        resources:
          requests:
            cpu: "100"
            memory: 200Gi

As any node of member clusters does not have so many cpu and memory resources, we will find workload scheduling failed.

$ kubectl get deployments.apps 
NAME    READY   UP-TO-DATE   AVAILABLE   AGE
nginx   0/5     0            0           2m20s
$ kubectl get rb nginx-deployment -o=custom-columns=NAME:.metadata.name,CLUSTER:.spec.clusters  
NAME               CLUSTER
nginx-deployment   <none>